Detection of Crack on Asphalt Pavement using Deep Convolutional Neural Network

Detection of crack on asphalt pavement is an essential task of monitoring and regulatory inspection. Currently, this task is conducted manually by surveyor or human inspectors for further maintenance works. Manual practice would lead some drawback such as time-consuming, labour intensive, hazardous and also subjective valuation for different individual. To overcome this deficit circumstances an automated technique is implemented. The objective of this study is to develop an intelligent system to detect pavement crack using Deep Convolutional Neural Network (DCNN). This study consists of several procedures which is started with collecting pavement crack images using online and from own developed dataset. The images are pre-processed by resizing the image into desire dimensions. Next, small patches are extracted as inputs to ease of detection and reduce classifier burden. The images further be labelled into two (2) types which is crack and non-crack. In this study, it is utilized Python environment and Keras framework to establish DCNN model. 80% of dataset is used for training set to train, while another 20% is used for testing set to test the model in order to evaluate the performance in terms of accuracy, precision and recall and F1 score. This proposed model is compared on different patch sizes, training algorithms and architectures to get the best classification. Thus, an automated system that able to accurately detect the present of crack in pavement images within speedy computation is successfully developed. To conclude, the system can be used to assist the surveyor or human operator in task of crack detection, so that the process of detection can be done faster and more efficient. This will help in reducing cost of maintenance and enhancing safety of road users. (paper)